SteadySplats: Resampling of Low-Variance Gaussians for High-Fidelity Stochastic Rendering

2026-10-04

AI Summary

This study addresses the high-frequency noise problem in 3D Gaussian Splatting caused by order-independent transparency rendering with stochastic sampling. To mitigate this, we jointly optimize the stochastic rendering process at both the representation and compositing layers. Specifically, we propose a history-based spatial resampling strategy to accelerate convergence, combined with temporal importance resampling to maintain coherence during camera motion, and introduce color regularization to suppress variance. Furthermore, an efficient renderer is implemented using the Vulkan API. Experimental results demonstrate that our method achieves a 13 dB PSNR improvement under single-sample-per-pixel rendering, rapidly converging to the quality of sorted 3DGS with an average L1 error below 10⁻⁴, thereby enabling high-quality real-time rendering.

Abstract

Stochastic order-independent transparency enables efficient and elegant rendering of primitive-based radiance fields like 3D Gaussian Splatting models, but remains impractical due to the inherent visible noise in the output. We propose a principled approach to minimize high-frequency noise, addressing its sources at the representation and image synthesis level. During stochastic rendering, our history-based spatial resampling scheme drastically accelerates image convergence, while temporal importance resampling ensures coherence under camera movement. During training, a color regularizer implicitly reduces the variance along view rays in the 3DGS models. With these properties, our optimized, Vulkan-based renderer effectively mitigates output noise at low and high sample counts, achieving a substantial 13~dB PSNR increase in quality over previous stochastic methods at 1 sample per pixel and quickly converging to sorted 3DGS with an average L1 error of less than $10^{-4}$.

Problem

• Stochastic Rendering
• Order-Independent Transparency
• 3D Gaussian Splatting
• Noise Reduction
• Radiance Fields

Innovation

• Stochastic Rendering
• 3D Gaussian Splatting
• Spatial-Temporal Resampling
• Order-Independent Transparency
• Color Regularization